Here's a basic rundown of ATI in the USA:
1) Marlborough MA : Home of ATI Design Team East. Previously developed core for R100 & R200. Currently developing some new core. Also home base of OpenGL driver developement. The cool technology demoes are made here too.
2) Silicon Valley CA : Home of ATI Design Team West. Formerly ArtX. Previously developed core for R300. Currently developing some new core. Also home base of R300 D3D developement.
3) Orlando FL : ATI's research campus. Formerly Real3D. Is principal contributor to future 3d development such as new pixel/vertex shader specs, DX research, OGL2 and so forth. Dabbles in ATI's drivers as well.
You can basically figure out what each location at ATI does by looking at the available job postings on their website, but I also know a little bit more information from friends As you can see, they get a lot of the "cool" jobs.
ATI in Canada:
Corporate headquarters. Marketing headquarters. OEM support/relationship headquarters. Board engineering and design. Engineering diagnostics labs. Small board manufacturing plant. 2D drivers. R100/R200 D3D drivers. Multimedia drivers / everything AIW. Driver packaging/installation/control panel. Catalyst program. Internal driver testing and WHQL testing. And lastly, everything else that ATI does that does not belong in the discrete graphics space (ie set-top box, PDA, etc).
Concerning ArtX's contribution, it was big but they did not design R300 from scratch. There is a great deal of implementation that was carried over from R200 to R300 and improved upon: The aniso algorithm. The occlusion detection algorithms. DX8.1 pixel shaders (ATI always said that the R200's pixel shaders were a prelude to DX9). R200's AA sample patterns (also programmable before fog bug squashed it). The 2D/overlay portion of the asic was also based on R200.
That said, I have no idea who was responsible for making the chip with 50 million more transistors operate at 400mhz+ without a process shrink versus R200
